QENEX LAB v2 · ChemRxiv 2026-05-26

Run quantum chemistry.
Ship cryptographic provenance.

For UK chemistry & materials groups. Pull the Apache-2.0 verifier subset, run any of 12 quantum chemistry methods against six basis sets, and emit a runnable .qlang verification program plus an atoms.json sidecar with every accepted output. Reviewers reproduce every claim on a clean machine in under fifteen minutes. UKAMF identity binding for federated researcher authentication.

Zenodo DOI 10.5281/zenodo.20083483 Apache License 2.0 UK Companies House 16523814

Run a calculation. See the numbers.

paper Tables I & II · live
// QENEX LAB v2 // values from §IV
0
user-callable methods
0 / 6
sub-nano tuples vs PySCF
0 / 113
self-assessed compliance
0
automated tests
// Computational engine

Twelve methods. One unified API.

All accessible through Molecule.compute(method=…). Six basis sets: STO-3G, 6-31G*, cc-pVDZ, cc-pVTZ, aug-cc-pVDZ, aug-cc-pVTZ. Pure-Python with optional libcint acceleration.

// 01
HF
Hartree–Fock (RHF) with DIIS + adaptive start-up
// 02
UHF
Unrestricted HF with SAD guess
// 03
MP2
Second-order Møller–Plesset, frozen-core option
// 04
CCSD
Coupled cluster singles & doubles, Stanton conventions
// 05
CCSD(T)
Perturbative triples correction
// 06
UCCSD
Open-shell CCSD in spin-orbital basis
// 07
EOM-CCSD
Equation-of-motion CCSD, Davidson iteration
// 08
CASSCF
Complete-active-space SCF with full CI
// 09
DFT
Kohn–Sham with LDA + B3LYP, Becke partitioning
// 10
TDDFT
Linear response (TDA + full RPA)
// 11
DLPNO-CCSD
Domain-based local-pair natural-orbital CCSD
// 12
DMC
Diffusion Monte Carlo
// Numerical agreement vs PySCF 2.4

Sub-nanohartree on 5 of 6 tuples.

Hartree–Fock total energies compared to PySCF reference values. The libcint path achieves sub-nanohartree on every tuple; the native Obara–Saika engine is sub-nanohartree on s-only systems and sub-millihartree on polyatomics — a documented limitation of the p-shell recursion (paper §IV.A).

Native path drift

absolute Δ vs PySCF · log scale
SystemQENEXPySCFlibcint Δ
He/STO-3G−2.8077839575−2.8077839575<10⁻¹²
H₂/STO-3G−1.1167593074−1.1167593074<10⁻¹⁰
H₂O/STO-3G−74.9630231385−74.9630231385<10⁻¹⁰
He/cc-pVDZ−2.8551604772−2.8551604772<10⁻¹²
H₂/cc-pVDZ−1.1287094490−1.1287094490<10⁻¹²
H₂O/cc-pVDZ−76.0271401825−76.0271401825<10⁻¹⁰
Two consecutive runs of precision_matrix.py --json produce bit-identical output (verified by SHA-256 against MANIFEST.sha256).
// Self-assessed posture · §VII

111 of 113 criteria. Six published rubrics.

Internally audited against six rubrics, with the audit script (standards_audit.py) producing bit-identical output across consecutive runs. These are self-assessments, not external audits by NASA, DoD, ISO, IEC, or WIPO — the paper documents this distinction explicitly.

6
/ 9
// NASA
Technology Readiness Level
System pinned to minimum per NASA SP-20170005794. drug_discovery domain at 7.
7
/ 10
// DoD
Manufacturing Readiness Level
13/13 criteria PASS. Honest dual-bible alignment.
40
/ 42
// NASA NPR 7150.2D
Software Engineering
38 PASS + 2 honest WARN (SWE-005, SWE-205).
20
/ 20
// FAIR4RS v1.0
FAIR Principles · Research Software
Findable · Accessible · Interoperable · Reusable.
18
/ 18
// ISO/IEC 25010:2011
Software Quality Model
All eight quality characteristics satisfied.
8
/ 8
// WIPO
PCT Rule 11 · Patent Format
International search and preliminary examination ready.
111 / 113
Aggregate · self-assessed · reproducible
A reviewer who disagrees can re-run standards_audit.py and inspect the evidence chain for every criterion.
// Apache 2.0 · §VI · ~15 minutes

Quickstart. Run the verifier.

Clone the open-source verifier subset, install, run. Two consecutive runs produce bit-identical output verified by SHA-256 against MANIFEST.sha256. No network access, no MCP tool, no proprietary lab modules required.

$ git clone https://github.com/qenex-ai/qenex-verifier.git
$ cd qenex-verifier
$ pip install -e ".[test,pyscf]"
$ bash verify.sh
 
[1/4] method_inventory.py → 12 user-callable methods ✓ SHA-256 match
[2/4] module_inventory.py → 13 verifier / 32 lab modules ✓ SHA-256 match
[3/4] qlang_inventory.py → 16 modules / 5,593 lines ✓ SHA-256 match
[4/4] precision_matrix.py → 5 / 6 tuples sub-nanohartree ✓ SHA-256 match
 
$ bash verifier_evaluator.py full
chemistry validation suite → 22 / 22 ZERO DEFECTS
 
VERIFY OK — every numeric claim in the v2 abstract reproduces bit-for-bit.
runtime: 11m 38s on a stock CPython 3.10 + NumPy + SciPy + Numba install.
// Cryptographic provenance · §VIII.D

Every output ships with a runnable proof.

SHA-256 Merkle chain over input geometry, integral checksums, SCF convergence trace, and final energies. Each accepted discovery emits a .qlang verification program plus an atoms.json sidecar that an external auditor can re-execute.

atoms.json
Input geometry · Bohr units
.qlang
Runnable verification program
Merkle root
SHA-256 over computation chain
External replay
Bit-identical reproduction
// For UKAMF reviewers + procurement

SAML SP identity. Federation metadata.

QENEX Lab acts as a SAML Service Provider in the UK Access Management Federation (UKAMF / Jisc). Reviewer-facing identity details:

PropertyValue
SAML SP entityIDhttps://lab.qenex.ai/sp
Assertion Consumer Servicehttps://auth.qenex.ai/realms/qenex/broker/ukamf/endpoint
NameID formaturn:oasis:names:tc:SAML:2.0:nameid-format:persistent
SAML bindingHTTP-POST
SP signing certECDSA P-256 · valid 2026-05-19 → 2028-05-18
SP cert SHA-256F9:BD:1F:69:9B:A3:59:06:3F:68:FB:FC:A9:74:DA:E9:9D:56:65:6F:6F:AF:52:67:85:7C:BC:10:11:2E:EF:4F
UKAMF membershipMember confirmed by Jisc; registration of lab.qenex.ai SP unblocked, awaiting DNS TXT validation token.
Attribute release expectededuPersonScopedAffiliation, mail (required); eduPersonPrincipalName, cn, eduPersonOrcid (optional)

SAML SP metadata XML on request to ceo@qenex.ai; UKAMF distributes via the federation hub on approval. UKAMF technical contact: admin@qenex.ai.

// Output structure

What every Lab output asserts on its receipt.

Each .qlang verification program + atoms.json sidecar carries a JSON receipt whose every field is independently checkable. The receipt does not depend on trusting QENEX:

FieldWhat it assertsVerifiable without trusting QENEX
artifact_sha256The hash of the canonical bytes of the computation outputBit-level — recompute the SHA-256 yourself
operator_signatureEd25519 signature by the QENEX operator key over artifact_sha256Verify against the published QENEX public key
prev_chain_hashHash of the previous accepted receipt — forms a tamper-evident chainWalk the chain; any break invalidates the lineage
tsa_proofRFC 3161 timestamp from Belgian BOSA qualified TSA (EU Trust List)Verify with openssl ts -verify + BOSA root certificate
ots_proofOpenTimestamps Bitcoin anchorVerify against any Bitcoin full node
rekor_uuidSigstore Rekor private transparency-log entryRead from rekor.qenex.ai — tamper-evident log
identitiesResearcher identities released by the UKAMF IdP (EPPN, ORCID, affiliation)Submitter authenticated to a UKAMF-member institution at the moment of anchoring
// For first-paid-pilot PIs

Researcher workflow. Three steps.

// Step 1
Authenticate
Via your institutional identity provider through the UK Access Management Federation. One-click flow from your institution's IdP discovery page after UKAMF approval.
// Step 2
Compute & submit
Run any of the 12 methods through Molecule.compute(). The platform emits the .qlang verification program plus atoms.json geometry sidecar automatically.
// Step 3
Anchor & share
Receipt is signed (Ed25519), timestamped (BOSA + OpenTimestamps), and chained. Share the receipt URL with reviewers — they verify offline against the public TSA certificates.

Method-level detail in the paper's §VIII. Open-source verifier: qenex-ai/qenex-verifier (Apache 2.0).

// Public record

Milestones on the public record

22 May 2026
SVAI/56 — £2M DSIT Sovereign AI Strategic Assets grant proposal submitted. 12-month programme funds the UK-Governed Open Benchmark Dataset for Reverse Discovery plus three UK university pilots.
26 May 2026
ChemRxiv v2 preprint live. 12 methods, 6 basis sets, sub-nanohartree precision on 5/6 tuples, 111/113 self-assessed compliance. Supersedes v1 with honest reconciliation. Verifier on GitHub + Zenodo (10.5281/zenodo.20083483).
10 Jun 2026
WIPO Madrid trademark filed (basic UK00004397356, EU+US designations). Priority locked.
15 Jun 2026
AIRR Grace Hopper Rapid Access proposal submitted to UKRI. Sovereign UK compute lane.
QENEX Pulse is archived. The cryptographic third-party-risk product is no longer marketed; QENEX is focused on the Lab. pulse.qenex.ai stays reachable for in-flight users; existing customers can contact ceo@qenex.ai for migration support.